feat(skills): add production media generation

[Catalina-df]

Summary

• add a production-ready media-generation skill for the configured new-api endpoint
• provide doctor, models, image, and video commands with progressive references
• use OPENAI_COMPATIBLE_BASE_URL and OPENAI_COMPATIBLE_API_KEY only
• bundle AI SDK dependencies into a portable 508 KiB executable JavaScript artifact
• validate model capabilities, local media, output collisions, timeouts, error redaction, and bounded downloads

Verification

• source CLI: 9 passed, 0 failed (1 bundle-only test skipped)
• bundled CLI: 10 passed, 0 failed
• bundle runs with skill-local node_modules temporarily unavailable
• typecheck, format, git diff --check, and secret scan passed
• live gpt-image-2 call through axolotl new-api succeeded and produced a valid PNG

Commit: db6d63a

[DF-wu]

Reviewed head 07bff5c from the DF-wu account on axolotl.

Verified:

• the skill is indexed and declares its Node runtime requirement
• credentials remain env-only and provider diagnostics redact the API key
• gpt-image-2 request mapping, capability validation, no fallback/retries, exclusive output handling, and the portable bundle
• source tests 9/9, bundle tests 10/10, typecheck/format/diff checks, and bun audit with no vulnerabilities
• a live gpt-image-2 request through the axolotl new-api endpoint produced a valid PNG

Video remains covered by protocol boundary tests and is not a live acceptance gate per the owner’s instruction. No blocking findings.

[DF-wu]

Re-reviewed latest head 64bc736 after the independent review pass.

The valid follow-up finding was fixed: terminal video task errors now pass through credential redaction, with source and bundled CLI regression coverage. Runtime metadata and the active skill index are also present.

Latest verification: source 10/10, bundle 11/11, typecheck/format/diff checks, bun audit clean, and the prior live gpt-image-2 production call succeeded. No blocking findings on the latest head.